Knowledge Support for Emergency Response During Construction Safety Accidents

Emergency response to construction safety accidents is the focus of this study. Despite the abundance of data and materials available for emergency response in construction safety, the unstructured nature of the knowledge and the disordered state of storage have limited the timely application of this knowledge in decision-making for emergency response. In this study, scenario-response theory, natural language processing, and deep learning technologies were employed to construct a domain knowledge graph for emergency response in the field of safety accidents. First, based on scenario-response theory and domain-specific materials, four categories of scenario domains and 14 types of scenario elements were identified. Second, according to the mapping relationships between scenario elements and emergency response knowledge, 14 entity types and 10 relationship types were determined, thereby forming the knowledge structure pattern of this field. Subsequently, 4877 entities and 5783 relationships were extracted by means of the BERT-BiLSTM-CRF model and the BERT-CNN model, with F1 values reaching approximately 0.8. Finally, the Neo4j graph database was adopted for data storage, and a domain knowledge graph was constructed. Based on this graph, services such as knowledge association, knowledge retrieval, and intelligent question-answering were implemented. These services effectively addressed key challenges in information acquisition and decision support for on-site safety management, thereby significantly enhancing response efficiency and quality while strengthening overall safety management practices within the construction industry.